Deep learning optimization for small object classification in lensfree holographic microscopy.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Optics and Photonics

Background:

• Lensfree holographic microscopy offers high-resolution, large-field-of-view imaging.
• Automated image processing with neural networks enhances biomolecular sensing using labeled micro- and nano-beads.
• Optimal neural network architectures for small object classification in holographic microscopy remain underexplored.

Purpose of the Study:

• To investigate the performance of shallow convolutional neural networks for small object classification in lensfree holographic microscopy.
• To analyze the impact of various network layers and hyperparameters on classification accuracy.

Main Methods:

• Application of a shallow convolutional neural network to classify small objects in holographic microscopy images.
• Systematic evaluation of layers (dropout, convolutional, normalization, pooling, activation) and hyperparameters (dropout fraction, filter number/size, stride, padding).

Main Results:

• Achieved a classification accuracy of approximately 83%.
• Identified the activation layer as the most critical factor for maximizing accuracy.
• Demonstrated the effectiveness of convolutional neural networks for this specific classification task.

Conclusions:

• Shallow convolutional neural networks are suitable for small object classification in lensfree holographic microscopy.
• Careful selection of network architecture, particularly the activation layer, is crucial for optimizing sensor performance.
• Findings provide guidance for developing neural networks in similar holographic-based sensing applications.